agairola/frame-tv

Frame TV Art Skill

Generate AI artwork or resize existing images for Samsung Frame TV displays. Uses Google Gemini (nano-banana-2 style) for image generation and Pillow for precision resizing.

Install

npx skills add agairola/life-skills --skill frame-tv

When to Use

Trigger this skill when the user:

• Asks to generate art for a Samsung Frame TV
• Wants to create wall art or display art using AI
• Asks to resize an image for their Frame TV
• Mentions Samsung Frame TV art or TV art generation
• Wants AI-generated artwork for a specific TV size
• Asks about Frame TV resolutions or image sizing
• Wants to create gallery-quality images for display

Prerequisites

• uv — brew install uv (macOS) or pip install uv (all platforms)
• API keys — GEMINI_API_KEY required for image generation (free at Google AI Studio). Not needed for resize-only mode.
• Dependencies — declared inline (PEP 723), installed automatically by uv run.

See references/api_upgrade.md for API key setup instructions.

Setup Status

!command -v uv > /dev/null 2>&1 && echo "uv: installed" || echo "uv: NOT INSTALLED" !test -n "${GEMINI_API_KEY:-}" && echo "GEMINI_API_KEY: set" || (test -f ~/.nano-banana/.env && echo "GEMINI_API_KEY: found in ~/.nano-banana/.env") || echo "GEMINI_API_KEY: NOT SET (needed for generation, not resize)"

Command Template

uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" [OPTIONS]

Options

| Flag | Values | Default | Purpose | |------|--------|---------|---------| | --prompt / -p | text | (none) | Text prompt for AI image generation | | --resize / -r | file path | (none) | Resize an existing image for Frame TV | | --input-image / -i | file path(s) | (none) | Reference image(s) for generation (up to 14, repeatable) | | --tv / -t | 32, 43, 50, 55, 65, 75, 85 | 55 | Samsung Frame TV size in inches | | --resolution | 512px, 1K, 2K, 4K | 4K | Generation resolution preset (auto-detected from input image if not set) | | --output-dir / -o | directory path | art | Output directory | | --model / -m | Gemini model ID | gemini-3.1-flash-image-preview | Gemini model to use | | --aspect / -a | ratio (e.g., 16:9) | 16:9 | Aspect ratio (passed via API image_config) | | --api-key | API key string | (none) | Gemini API key (overrides env vars) | | --preview | (flag) | off | Generate a quick 512px preview (use --upscale to finalize) | | --upscale | file path | (none) | Upscale a preview image to full 4K resolution | | --dry-run | (flag) | off | Validate without API call |

Only parse stdout (JSON). Stderr contains diagnostics only.

Common Commands

# Preview first, then upscale (RECOMMENDED workflow)
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "calm ocean sunset, oil painting style" --preview
# → shows 512px preview, user reviews it
# → if approved, run the upscale_command from the JSON output:
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --upscale art/frame_art_<timestamp>_raw.png

# Generate directly at 4K (skip preview)
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "calm ocean sunset, oil painting style"

# Generate for a specific TV size
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "Japanese garden in autumn" --tv 65

# Resize an existing image for Frame TV
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --resize /path/to/image.jpg --tv 55

# Use a reference image for style/content guidance
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "in this style but with autumn colors" --input-image ref.jpg

# Multiple reference images
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "combine these styles" --input-image style1.jpg --input-image style2.jpg

# Square aspect ratio (for specific art styles)
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "abstract geometric art" --aspect 1:1

# Specify resolution preset
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "landscape painting" --resolution 2K

# Pass API key directly
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "ocean sunset" --api-key "your-key-here"

# Dry run to check config
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --prompt "test" --dry-run

# Resize for a smaller 32" TV (1080p)
uv run "${CLAUDE_SKILL_DIR}/scripts/frame_tv_art.py" --resize photo.jpg --tv 32

Presenting Results

Follow the formatting rules in ../../references/platform-formatting.md. Key skill-specific formatting below.

Preview Mode (recommended default)

Frame TV Art Preview:

Prompt: "calm ocean sunset, oil painting style"
Preview: art/frame_art_20260319_143022_raw.png (512px)

Does this look good? If yes, I'll generate the full 4K version for your 55" TV.

When the user approves, run the upscale_command from the JSON output. After upscale:

Frame TV Art — Full Resolution:

Prompt: "calm ocean sunset, oil painting style"
TV Size: 55" (3840×2160)
Upscaled from preview → 3840×2160
Output: art/frame_art_55in_20260319_143122.jpg

Transfer tip: Copy to USB drive for best quality — avoids compression from phone/cloud transfer.

Generate Mode (direct, no preview)

Frame TV Art Generated:

Prompt: "calm ocean sunset, oil painting style"
TV Size: 55" (3840×2160)
Generated: 1024×1024 → resized to 3840×2160
Output: art/frame_art_55in_20260319_143022.jpg

Transfer tip: Copy to USB drive for best quality — avoids compression from phone/cloud transfer.

Resize Mode

Image Resized for Frame TV:

Input: vacation_photo.jpg (4000×3000)
TV Size: 55" (3840×2160)
Output: art/vacation_photo_frame_55in_20260319_143022.jpg (3840×2160)

Transfer tip: Copy to USB drive for best quality.

Formatting Rules

• Default to --preview mode for generation — show the preview and ask the user before upscaling to 4K
• Always show the TV size and target resolution
• Show the resize dimensions (original → final)
• Include the output file path
• Always mention the USB transfer tip for best quality (after upscale/generate, not for previews)
• If API key is missing, show setup instructions from references/api_upgrade.md
• For resize-only mode, note that no API key is needed

Handling Edge Cases

• No API key (generate mode): Show the error with setup instructions — "Set GEMINI_API_KEY or create ~/.nano-banana/.env. Get a free key at aistudio.google.com/apikey"
• API key present but generation fails: Show the error message from Gemini. Suggest trying a different prompt or checking API quota.
• Resize input not found: "File not found: [path]. Check the file path and try again."
• No --prompt or --resize: "Either --prompt or --resize is required. Use --prompt for AI generation or --resize to fit an existing image."
• Unsupported TV size: argparse handles this — shows valid choices (32, 43, 50, 55, 65, 75, 85).

Reference

Samsung Frame TV Resolutions

| TV Size | Resolution | |---------|-----------| | 32" | 1920×1080 (Full HD) | | 43" | 3840×2160 (4K UHD) | | 50" | 3840×2160 (4K UHD) | | 55" | 3840×2160 (4K UHD) | | 65" | 3840×2160 (4K UHD) | | 75" | 3840×2160 (4K UHD) | | 85" | 3840×2160 (4K UHD) |

Art Style Tips

• Oil paintings: Rich textures, vibrant colors — great for living rooms
• Watercolors: Soft, light feel — works well in bedrooms
• Photography: Nature scenes, cityscapes — versatile for any room
• Abstract: Bold patterns and colors — modern/contemporary spaces
• Add "museum-quality" or "gallery-worthy" to prompts for higher quality results

Image Transfer

For best quality on your Frame TV:

1. Save images to a USB drive (FAT32 or exFAT formatted)
2. Insert USB into TV's USB port
3. Open Art Mode → My Collection → USB
4. Select images to import

Avoid transferring via messaging apps (WhatsApp, iMessage) — they compress images significantly.

Technology

Image generation uses Google Gemini (nano-banana-2 approach) with gemini-3.1-flash-image-preview. Aspect ratio and resolution are passed via image_config in the API config (not just prompt text). Supports reference/input images sent alongside the prompt. Handles multiple output images from a single generation. RGBA images are converted to RGB before JPEG saving. Images are resized to exact TV dimensions using Pillow's LANCZOS resampling. Output is high-quality JPEG (quality=92, no chroma subsampling).